Multitask genetic programming for automated design of heuristics for the container relocation problem

Created by W.Langdon from gp-bibliography.bib Revision:1.8444

@Article{Durasevic:2025:engappai,

• author = "Marko Durasevic and Mateja Dumic and Francisco Javier {Gil Gala}",
• title = "Multitask genetic programming for automated design of heuristics for the container relocation problem",
• journal = "Engineering Applications of Artificial Intelligence",
• year = "2025",
• volume = "144",
• pages = "110001",
• keywords = "genetic algorithms, genetic programming, Multitask genetic programming, Container relocation problem, Hyper-heuristics, Relocation rules",
• ISSN = "0952-1976",
• URL = "https://www.sciencedirect.com/science/article/pii/S0952197625000016",
• DOI = "doi:10.1016/j.engappai.2025.110001",
• abstract = "Automated design of heuristics with genetic programming (GP) received significant attention for various optimisation problems. One of these is the container relocation problem (CRP), encountered in shipping terminals and warehouses. Although GP can design high quality heuristics, it has certain limitations. One is that numerous variants of each optimisation problem exist, meaning that appropriate heuristics have to be designed for each. Most commonly these heuristics are evolved independently from each other, which can be regarded as wasteful since they should exhibit a certain kind of similarity because they solve the same problem. This encouraged the development of knowledge transfer methods, such as multitask learning, that share the knowledge between solutions evolved for different problem types. In multitask learning several problems are solved simultaneously, with solutions of different problems shared between the populations that solve each problem. Until now, in the context of the automated design of heuristics, multitask learning has been primarily used to solve scheduling problems with a limited number of different problem types. This motivates us to investigate multitask learning with two main objectives. First, to apply multitask learning to evolve heuristics for a different problem, namely CRP. Second, to investigate how problem types that are optimised simultaneously influence its ability to obtain efficient heuristics. The obtained results demonstrate that multitask learning improves the convergence of GP, although the overall performance of the obtained heuristics remains similar. Furthermore, the effectiveness of multitask learning significantly depends on the problem types optimised simultaneously, outlining that the choice of the problems is critical",

}

Genetic Programming entries for Marko Durasevic Mateja Dumic Francisco Javier Gil Gala

Citations